Skip to content

Sarvam 105B 1711

Sarvam's 105B MoE flagship joins the catalog on the new 1711 weights: 10.3B active parameters, a 128K context, tool calling and Indian-language support. BYOK only.

Open Sarvam 105B
Sarvam-105B-1711 hero art

sarvam/sarvam-105b runs Sarvam's 1711 weights. It is a 105B-parameter sparse mixture-of-experts model with 10.3B active parameters and multi-head latent attention, built for reasoning, agentic tool use and the ten most-spoken Indian languages.

The API exposes a 128K context window. Reasoning is on by default and reasoning tokens count toward max_tokens; set reasoning_effort to none to turn it off. Open weights are Apache 2.0 in NVFP4 and BF16.

AI SDK
import { createAnyRouter } from "@anyr/ai-sdk-provider"
import { generateText } from "ai"
const anyrouter = createAnyRouter()
const { text } = await generateText({ model: anyrouter("sarvam/sarvam-105b"), prompt: "Hello" })
Coding agent
curl -fsSL https://anyr.dev/cli.sh | bash
anyr login
anyr claude --model "sarvam/sarvam-105b"

It is BYOK-only. Add your own Sarvam key under Dashboard → BYOK. AnyRouter does not charge credits for the hop. Read the model card for the full benchmark table.

Sarvam-105B-1711 benchmark-average versus blended price chart
Vendor chart: 73.9 benchmark average at about $0.15 per 1M blended tokens, compared with Qwen3.8 27B, Gemini 3.5 Flash-Lite and others.. Source: Sarvam AI.

Looking for the full catalog? Browse all models.